Multifidelity data fusion method for electromagnetic field prediction of linear propulsion electromagnetic energy equipment

Abstract

The application provides a multi-fidelity data fusion method for predicting an electromagnetic field of linear propulsion electromagnetic energy equipment, comprising: constructing a multi-fidelity data set about the electromagnetic field of linear propulsion electromagnetic energy equipment, wherein the data set comprises low-fidelity data and high-fidelity data; based on a set knowledge distillation framework, combining a deep transfer learning model and a loss function reconstruction strategy to perform multi-fidelity data fusion; wherein the deep transfer learning model comprises a teacher model and a student model, the teacher model is trained based on the low-fidelity data, the student model is trained based on the high-fidelity data, and the structure backbone in the teacher model is transferred to the student model. The data fusion method can effectively combine the low-fidelity data with low calculation time and the high-fidelity data with high precision, and realize the optimal trade-off between calculation time and prediction accuracy under limited resources.

Description

Technical Field

[0001] This application belongs to the field of electromagnetic field distribution prediction technology, and in particular relates to a multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment. Background Technology

[0002] Currently, data-driven deep learning models can accurately predict the electromagnetic field distribution of linear propulsion electromagnetic energy thanks to their ability to capture multi-scale features and enhance feature extraction capabilities, but they require a large amount of high-fidelity electromagnetic field data.

[0003] However, due to the need for high-precision simulations that linearly advance electromagnetic energy, the acquisition of high-fidelity magnetic field data is often expensive. On the other hand, low-fidelity magnetic field data obtained through dimensionality reduction, model simplification, or coarse numerical simulation are significantly cheaper to acquire and can still provide useful information on electromagnetic field trends, but they suffer from lower accuracy. Summary of the Invention

[0004] In view of this, this application aims to propose a multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment, so as to solve at least one of the above problems.

[0005] To achieve the above objectives, the technical solution of this application is implemented as follows:

[0006] Firstly, this application provides a multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment, including:

[0007] Construct a multi-fidelity dataset of electromagnetic fields for linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data;

[0008] Based on the established knowledge distillation framework, and combined with a deep transfer learning model and a loss function reconstruction strategy, multi-fidelity data fusion is performed; wherein, the deep transfer learning model includes a teacher model and a student model, the teacher model is trained based on the low-fidelity data, and the student model is trained based on the high-fidelity data, and the structural backbone of the teacher model is transferred to the student model.

[0009] Secondly, based on the same inventive concept, this application also provides a multi-fidelity data fusion device for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment, comprising:

[0010] The dataset construction module is configured to construct a multi-fidelity dataset about the electromagnetic field of a linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data;

[0011] The data fusion module is configured to perform multi-fidelity data fusion based on a set knowledge distillation framework and combined with a deep transfer learning model and a loss function reconstruction strategy. The deep transfer learning model includes a teacher model and a student model. The teacher model is trained based on the low-fidelity data, and the student model is trained based on the high-fidelity data. The structural backbone of the teacher model is transferred to the student model.

[0012] Thirdly, based on the same inventive concept, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0013] Fourthly, based on the same inventive concept, this application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions for causing the computer to perform the method as described in the first aspect.

[0014] Compared with existing technologies, the multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment described in this application has the following advantages:

[0015] The multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment described in this application aims to reduce the computation time required to construct the dataset for the electromagnetic field prediction task of linear propulsion electromagnetic energy equipment. It uses a knowledge distillation framework based on a set condition and combines a deep transfer learning model and a loss function reconstruction strategy to perform multi-fidelity data fusion. This method can effectively combine low-fidelity data with low computation time and high-fidelity data with high precision, achieving the optimal trade-off between computation time and prediction accuracy with limited resources. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a flowchart of the multi-fidelity data fusion method for predicting the electromagnetic field of linear propulsion electromagnetic energy equipment according to an embodiment of this application;

[0018] Figure 2 This is a schematic diagram illustrating the construction of the dataset described in an embodiment of this application;

[0019] Figure 3 This is a diagram of the knowledge distillation-based training framework described in the embodiments of this application;

[0020] Figure 4This is a schematic diagram of the deep transfer learning model described in the embodiments of this application;

[0021] Figure 5 This is a schematic diagram of the structural backbone and adaptation layer structure of the deep transfer learning model described in the embodiments of this application;

[0022] Figure 6 This is a schematic diagram of the loss function reconstruction described in an embodiment of this application;

[0023] Figure 7 This is a schematic diagram of the multi-fidelity data fusion device for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment as described in an embodiment of this application.

[0024] Figure 8 This is a schematic diagram of the hardware structure of the electronic device described in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0026] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0027] The embodiments of this application are described in detail below with reference to the accompanying drawings.

[0028] Please see Figure 1 As shown, this embodiment provides a multi-fidelity data fusion method for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment, including:

[0029] Step S101: Construct a multi-fidelity dataset of the electromagnetic field of a linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data.

[0030] Specifically, in this embodiment, the multi-fidelity dataset construction process is as follows: Figure 2As shown, the steps are as follows: The optimal Latin hypercube sampling method (OLHS method) is used to select the geometric parameters of the linear propulsion electromagnetic energy equipment (dimensional parameters such as pivot length, contact surface size, and electromagnetic field distribution area size; shape parameters such as pivot shape, contact surface shape, and electromagnetic field distribution area shape; and relative position parameters such as the relative positions between pivots and between pivots and the propulsion device). Different mesh refinements for the pivot contact surfaces are selected to construct models with different fidelity, and the corresponding electromagnetic field data is extracted. In each sample, the input is a different geometric structure diagram, and the output is the corresponding electromagnetic field calculation result.

[0031] The exported dataset underwent data preprocessing, specifically as follows: First, the dataset was divided into training, validation, and test sets in an 8:1:1 ratio. The training set contained high-fidelity samples (β proportion) and low-fidelity samples (1-β proportion), while the validation and test sets contained only high-fidelity data. Finally, the magnetic field calculation results for different geometries were normalized to ensure all features were on the same scale, thus adapting to gradient-based optimization algorithms.

[0032] Step S102: Based on the established knowledge distillation framework, and combined with a deep transfer learning model and a loss function reconstruction strategy, multi-fidelity data fusion is performed; wherein, the deep transfer learning model includes a teacher model and a student model, the teacher model is trained on low-fidelity data, and the student model is trained on high-fidelity data, and the structural backbone in the teacher model is transferred to the student model.

[0033] Specifically, in this embodiment, the established knowledge distillation framework includes:

[0034] Applying knowledge distillation to multi-fidelity data fusion combines the advantages of loss function reconstruction and deep transfer learning. By adding a distillation loss function, knowledge distillation can simultaneously consider the different characteristics of low-fidelity and high-fidelity data and assign them appropriate weights to optimize the learning process. Furthermore, it borrows ideas from deep transfer learning, mimicking a sophisticated or better-performing teacher model to guide a simpler student model, thereby improving the student model's performance without significantly increasing computational costs.

[0035] In this process, the parameters of the student model are updated based on the comprehensive loss. The comprehensive loss consists of two parts: first, the difference between the outputs of the teacher model and the student model (i.e., distillation loss); and second, the difference between the student model output and the actual label (i.e., target loss). The formulas for calculating the distillation loss, target loss, and comprehensive loss are as follows:

[0036]

[0037]

[0038] L total =α·L target +(1-α)·L distillation (1.3)

[0039] In the formula, y t and y s y represents the output of the teacher model and the student model, respectively. y is the target true value, N is the number of data points, and α is a weight parameter between 0 and 1, which determines the contribution ratio of the target loss and distillation loss to the overall loss.

[0040] In the electromagnetic field prediction task of linear propulsion electromagnetic energy equipment, the knowledge distillation method can learn electromagnetic field trend information from low-fidelity data by leveraging the fitting ability of the teacher model. Subsequently, through knowledge distillation, the information learned from the teacher model is "distilled" into the student model to learn high-fidelity, detailed magnetic field information. The training framework is as follows: Figure 3 As shown, the training of the student model utilizes continuous value outputs provided by the teacher model. These continuous value outputs not only contain rich information but also reflect the teacher model's deep understanding of the data. Compared to traditional direct target value learning, this method enables the student model to reduce MSE loss while delving into the subtle connections between input and output, including the nonlinear properties of the data and complex output distribution patterns. This method effectively transforms the teacher model's profound insights into the data into the student model's predictive skills, thereby improving overall prediction accuracy.

[0041] like Figure 3 As shown, in the first stage, a large amount of low-fidelity data is used to train the teacher model to gain a deep understanding of the data. In the second stage, firstly, the parameters of the trained teacher model are frozen to ensure they remain unchanged during subsequent training; then, both the teacher model and the student model perform forward propagation on the same data to produce prediction results. In each training cycle, the student model performs backpropagation by minimizing the comprehensive loss function, updating only the student model's parameters, allowing the student model to both mimic the behavior of the teacher model and learn and improve directly from the real target data. After training, the effectiveness of the knowledge distillation method can be verified by evaluating the performance of the student model on the test set.

[0042] Multi-fidelity data fusion methods based on deep transfer learning, such as... Figure 4 As shown. First, the model is trained using a large amount of low-fidelity data, and the low-fidelity data features f are obtained through the structural backbone. L A low-fidelity model of the source domain is established; then, the structural backbone is transferred, the trained structural backbone parameters are frozen, and high-fidelity data (x) is used.H y H Continue training to obtain high-fidelity data features f H The weights of the high-fidelity adaptation layer are optimized. Finally, a global adjustment and optimization are performed, allowing all model parameters, including the structural backbone, to be adjusted to ensure that the model can fully learn and reflect the characteristics of the high-fidelity data.

[0043] This training process not only utilizes a large amount of low-fidelity data to capture basic features, but also uses high-fidelity data to fine-tune and optimize the model, ensuring high performance. Specifically, for the CI-UNet++ model, the division of the backbone and adaptation layers is as follows: Figure 5 As shown, the two 3×3 convolutional layers closest to the output serve as adaptation layers, while the rest form the structural backbone of the model.

[0044] In regression prediction problems, loss functions are typically considered only for high-fidelity data; however, multi-fidelity data fusion methods based on loss function reconstruction, such as... Figure 6 As shown, low-fidelity data can be included in the consideration with certain weights, thereby imposing a certain degree of constraint on the model to make more comprehensive use of all available data. The modified multifidelity loss function is:

[0045]

[0046] In the formula, N H and N L These represent the number of high-fidelity samples and low-fidelity samples, respectively. and These are the predicted values ​​for high-fidelity data and low-fidelity data, respectively. and 1 and 2 are the true values ​​of the high-fidelity data and the low-fidelity data, respectively; 1-k and k are the weighting coefficients of the high-fidelity data and the low-fidelity data, respectively.

[0047] Although low-fidelity data has lower accuracy, it serves as prior information, appropriately penalizing the loss function and compensating for the shortcomings of high-fidelity data; while high-fidelity data is used to further calibrate the model. This method can process data of different fidelities simultaneously and optimize the training process by adjusting the weights of these data in the loss function, helping the model maintain high predictive ability even when high-fidelity data is scarce.

[0048] When using a loss function reconstruction strategy, appropriately allocating weights for data of different fidelity levels is crucial for improving model prediction accuracy. In the electromagnetic field prediction task for linear propulsion electromagnetic energy equipment, given the high accuracy of high-fidelity data, it is assigned a larger initial weight to ensure that the model can effectively utilize this high-quality information. While low-fidelity data is abundant and covers a wide range, its accuracy is lower, so it is assigned a smaller initial weight.

[0049] To further optimize model performance, the Optuna hyperparameter optimization framework based on Bayesian optimization is used to dynamically adjust the weights. With Optuna's efficient search capabilities, the optimal weight coefficients can be found within a limited time.

[0050] When using this method for actual prediction, high-fidelity data is input into the trained student model, and the electromagnetic field prediction results of the linear propulsion electromagnetic energy equipment are output to complete the prediction of the electromagnetic field.

[0051] The fusion method described in this application, for the electromagnetic field prediction task of linear propulsion electromagnetic energy equipment, uses a knowledge distillation framework based on a set condition to reduce the computation time required to build the dataset. It combines a deep transfer learning model and a loss function reconstruction strategy to perform multi-fidelity data fusion, which can effectively combine low-fidelity data with low computation time and high-fidelity data with high precision. Under limited resources, it achieves the optimal trade-off between computation time and prediction accuracy.

[0052] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] Based on the same inventive concept, and corresponding to the methods of any of the above embodiments, the embodiments of this application also provide a multi-fidelity data fusion device for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment.

[0054] like Figure 7 As shown, the multi-fidelity data fusion device for predicting the electromagnetic field of the linear propulsion electromagnetic energy equipment includes:

[0055] The dataset construction module is configured to build a multi-fidelity dataset about the electromagnetic field of a linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data;

[0056] The data fusion module is configured to perform multi-fidelity data fusion based on a set knowledge distillation framework and combined with a deep transfer learning model and a loss function reconstruction strategy. The deep transfer learning model includes a teacher model and a student model. The teacher model is trained on low-fidelity data, and the student model is trained on high-fidelity data. The structural backbone of the teacher model is transferred to the student model.

[0057] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing the embodiments of this application, the functions of each module can be implemented in one or more software and / or hardware.

[0058] The apparatus of the above embodiments is used to implement the corresponding method in any of the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0059] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the methods described in any of the above embodiments.

[0060] Figure 8 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0061] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0062] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0063] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0064] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0065] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0066] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0067] The electronic devices described above are used to implement the corresponding methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0068] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to perform the methods described in any of the above embodiments.

[0069] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0070] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to perform the methods described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0071] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0072] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0073] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0074] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A multi-fidelity data fusion method for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment, characterized in that, include: Construct a multi-fidelity dataset of electromagnetic fields for linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data; Based on the established knowledge distillation framework, and combined with a deep transfer learning model and a loss function reconstruction strategy, multi-fidelity data fusion is performed; wherein, the deep transfer learning model includes a teacher model and a student model, the teacher model is trained based on the low-fidelity data, and the student model is trained based on the high-fidelity data, and the structural backbone of the teacher model is transferred to the student model; The knowledge distillation training framework includes: In the first stage, the teacher model is trained using the low-fidelity data. In the second stage, the parameters of the teacher model after training are frozen, and the teacher model and student model are respectively subjected to forward propagation operation on the same data to produce prediction results. In each training cycle, the student model performs backpropagation by minimizing the comprehensive loss function to update the parameters of the student model. After training is completed, the student model is evaluated using a test set.

2. The method according to claim 1, characterized in that, The construction of a multi-fidelity dataset regarding the electromagnetic field of a linear propulsion electromagnetic energy equipment includes low-fidelity data and high-fidelity data, comprising: The geometric parameters of the linear propulsion electromagnetic energy equipment are selected by the optimal Latin hypercube sampling method, and the mesh refinement of different pivot rail contact surfaces is selected to form models with different fidelity, and the corresponding electromagnetic field data is extracted. A multi-fidelity dataset is constructed based on the extracted electromagnetic field data, the dataset including low-fidelity data and high-fidelity data.

3. The method according to claim 2, characterized in that, Data preprocessing of the dataset includes: The dataset is divided into training, validation, and test sets according to a certain ratio; The training set contains high-fidelity data and low-fidelity data, and the validation set and the test set contain high-fidelity data. The sample data in the training set, the validation set, and the test set are normalized.

4. The method according to claim 1, characterized in that, The deep transfer learning model includes: The model is trained using the low-fidelity data, the low-fidelity data features are obtained through the structural backbone, and a low-fidelity model of the source domain is established. The structural backbone is transferred, the trained structural backbone parameters are frozen, and training is continued using the high-fidelity data to obtain high-fidelity data features and optimize the weights of the high-fidelity adaptation layer. The frozen structural backbone parameters are thawed, and then global adjustments and optimizations are performed.

5. The method according to claim 4, characterized in that, The loss function reconstruction strategy includes: Based on the actual and predicted values ​​of the low-fidelity data, and the actual and predicted values ​​of the high-fidelity data, the weights of each fidelity data are adjusted using a multi-fidelity loss function, which is: ； In the formula, N H and N L These represent the number of high-fidelity samples and low-fidelity samples, respectively. and These are the predicted values ​​for high-fidelity data and low-fidelity data, respectively. and These are the true values ​​of high-fidelity data and low-fidelity data, respectively; 1 -k and k These are the weighting coefficients for high-fidelity data and low-fidelity data, respectively.

6. The method according to claim 1, characterized in that, The parameters of the student model are updated based on the comprehensive loss, which is calculated using the following formula: ； ； ； In the formula, y t and y s y represents the output of the teacher model and the student model, respectively. y is the target true value, N is the number of data points, and α is a weight parameter between 0 and 1.

7. A multi-fidelity data fusion device for predicting the electromagnetic field of a linear propulsion electromagnetic energy equipment, characterized in that, include: The dataset construction module is configured to construct a multi-fidelity dataset about the electromagnetic field of a linear propulsion electromagnetic energy equipment, wherein the dataset includes low-fidelity data and high-fidelity data; The data fusion module is configured to perform multi-fidelity data fusion based on a set knowledge distillation framework and combined with a deep transfer learning model and a loss function reconstruction strategy; wherein, the deep transfer learning model includes a teacher model and a student model, the teacher model is trained based on the low-fidelity data, and the student model is trained based on the high-fidelity data, and the structural backbone of the teacher model is transferred to the student model; The knowledge distillation training framework includes: In the first stage, the teacher model is trained using the low-fidelity data. In the second stage, the parameters of the teacher model after training are frozen, and the teacher model and student model are respectively subjected to forward propagation operation on the same data to produce prediction results. In each training cycle, the student model performs backpropagation by minimizing the comprehensive loss function to update the parameters of the student model. After training is completed, the student model is evaluated using a test set.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, in, The non-transitory computer-readable storage medium stores computer instructions for causing a computer to perform the method described in any one of claims 1-6.

Images